changed reading hint
[gromacs/adressmacs.git] / src / fftw / fhb_8.c
blobb49baa5f4e00bbd5fb4d0905516ae74b2839d90d
1 /*
2 * Copyright (c) 1997-1999 Massachusetts Institute of Technology
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 /* This file was automatically generated --- DO NOT EDIT */
21 /* Generated on Tue May 18 13:56:04 EDT 1999 */
23 #include <fftw-int.h>
24 #include <fftw.h>
26 /* Generated by: ./genfft -magic-alignment-check -magic-twiddle-load-all -magic-variables 4 -magic-loopi -hc2hc-backward 8 */
29 * This function contains 108 FP additions, 50 FP multiplications,
30 * (or, 90 additions, 32 multiplications, 18 fused multiply/add),
31 * 31 stack variables, and 64 memory accesses
33 static const fftw_real K765366864 = FFTW_KONST(+0.765366864730179543456919968060797733522689125);
34 static const fftw_real K1_847759065 = FFTW_KONST(+1.847759065022573512256366378793576573644833252);
35 static const fftw_real K707106781 = FFTW_KONST(+0.707106781186547524400844362104849039284835938);
36 static const fftw_real K1_414213562 = FFTW_KONST(+1.414213562373095048801688724209698078569671875);
37 static const fftw_real K2_000000000 = FFTW_KONST(+2.000000000000000000000000000000000000000000000);
40 * Generator Id's :
41 * $Id$
42 * $Id$
43 * $Id$
46 void fftw_hc2hc_backward_8(fftw_real *A, const fftw_complex *W, int iostride, int m, int dist)
48 int i;
49 fftw_real *X;
50 fftw_real *Y;
51 X = A;
52 Y = A + (8 * iostride);
54 fftw_real tmp107;
55 fftw_real tmp118;
56 fftw_real tmp105;
57 fftw_real tmp116;
58 fftw_real tmp111;
59 fftw_real tmp120;
60 fftw_real tmp115;
61 fftw_real tmp121;
62 fftw_real tmp108;
63 fftw_real tmp112;
64 ASSERT_ALIGNED_DOUBLE();
66 fftw_real tmp106;
67 fftw_real tmp117;
68 fftw_real tmp103;
69 fftw_real tmp104;
70 ASSERT_ALIGNED_DOUBLE();
71 tmp106 = X[2 * iostride];
72 tmp107 = K2_000000000 * tmp106;
73 tmp117 = Y[-2 * iostride];
74 tmp118 = K2_000000000 * tmp117;
75 tmp103 = X[0];
76 tmp104 = X[4 * iostride];
77 tmp105 = tmp103 + tmp104;
78 tmp116 = tmp103 - tmp104;
80 fftw_real tmp109;
81 fftw_real tmp110;
82 fftw_real tmp113;
83 fftw_real tmp114;
84 ASSERT_ALIGNED_DOUBLE();
85 tmp109 = X[iostride];
86 tmp110 = X[3 * iostride];
87 tmp111 = K2_000000000 * (tmp109 + tmp110);
88 tmp120 = tmp109 - tmp110;
89 tmp113 = Y[-iostride];
90 tmp114 = Y[-3 * iostride];
91 tmp115 = K2_000000000 * (tmp113 - tmp114);
92 tmp121 = tmp114 + tmp113;
95 tmp108 = tmp105 + tmp107;
96 X[4 * iostride] = tmp108 - tmp111;
97 X[0] = tmp108 + tmp111;
98 tmp112 = tmp105 - tmp107;
99 X[6 * iostride] = tmp112 + tmp115;
100 X[2 * iostride] = tmp112 - tmp115;
102 fftw_real tmp119;
103 fftw_real tmp122;
104 fftw_real tmp123;
105 fftw_real tmp124;
106 ASSERT_ALIGNED_DOUBLE();
107 tmp119 = tmp116 - tmp118;
108 tmp122 = K1_414213562 * (tmp120 - tmp121);
109 X[5 * iostride] = tmp119 - tmp122;
110 X[iostride] = tmp119 + tmp122;
111 tmp123 = tmp116 + tmp118;
112 tmp124 = K1_414213562 * (tmp120 + tmp121);
113 X[3 * iostride] = tmp123 - tmp124;
114 X[7 * iostride] = tmp123 + tmp124;
117 X = X + dist;
118 Y = Y - dist;
119 for (i = 2; i < m; i = i + 2, X = X + dist, Y = Y - dist, W = W + 7) {
120 fftw_real tmp29;
121 fftw_real tmp60;
122 fftw_real tmp46;
123 fftw_real tmp56;
124 fftw_real tmp70;
125 fftw_real tmp96;
126 fftw_real tmp82;
127 fftw_real tmp92;
128 fftw_real tmp36;
129 fftw_real tmp57;
130 fftw_real tmp53;
131 fftw_real tmp61;
132 fftw_real tmp73;
133 fftw_real tmp83;
134 fftw_real tmp76;
135 fftw_real tmp84;
136 ASSERT_ALIGNED_DOUBLE();
138 fftw_real tmp25;
139 fftw_real tmp68;
140 fftw_real tmp42;
141 fftw_real tmp81;
142 fftw_real tmp28;
143 fftw_real tmp80;
144 fftw_real tmp45;
145 fftw_real tmp69;
146 ASSERT_ALIGNED_DOUBLE();
148 fftw_real tmp23;
149 fftw_real tmp24;
150 fftw_real tmp40;
151 fftw_real tmp41;
152 ASSERT_ALIGNED_DOUBLE();
153 tmp23 = X[0];
154 tmp24 = Y[-4 * iostride];
155 tmp25 = tmp23 + tmp24;
156 tmp68 = tmp23 - tmp24;
157 tmp40 = Y[0];
158 tmp41 = X[4 * iostride];
159 tmp42 = tmp40 - tmp41;
160 tmp81 = tmp40 + tmp41;
163 fftw_real tmp26;
164 fftw_real tmp27;
165 fftw_real tmp43;
166 fftw_real tmp44;
167 ASSERT_ALIGNED_DOUBLE();
168 tmp26 = X[2 * iostride];
169 tmp27 = Y[-6 * iostride];
170 tmp28 = tmp26 + tmp27;
171 tmp80 = tmp26 - tmp27;
172 tmp43 = Y[-2 * iostride];
173 tmp44 = X[6 * iostride];
174 tmp45 = tmp43 - tmp44;
175 tmp69 = tmp43 + tmp44;
177 tmp29 = tmp25 + tmp28;
178 tmp60 = tmp25 - tmp28;
179 tmp46 = tmp42 + tmp45;
180 tmp56 = tmp42 - tmp45;
181 tmp70 = tmp68 - tmp69;
182 tmp96 = tmp68 + tmp69;
183 tmp82 = tmp80 + tmp81;
184 tmp92 = tmp81 - tmp80;
187 fftw_real tmp32;
188 fftw_real tmp71;
189 fftw_real tmp49;
190 fftw_real tmp72;
191 fftw_real tmp35;
192 fftw_real tmp74;
193 fftw_real tmp52;
194 fftw_real tmp75;
195 ASSERT_ALIGNED_DOUBLE();
197 fftw_real tmp30;
198 fftw_real tmp31;
199 fftw_real tmp47;
200 fftw_real tmp48;
201 ASSERT_ALIGNED_DOUBLE();
202 tmp30 = X[iostride];
203 tmp31 = Y[-5 * iostride];
204 tmp32 = tmp30 + tmp31;
205 tmp71 = tmp30 - tmp31;
206 tmp47 = Y[-iostride];
207 tmp48 = X[5 * iostride];
208 tmp49 = tmp47 - tmp48;
209 tmp72 = tmp47 + tmp48;
212 fftw_real tmp33;
213 fftw_real tmp34;
214 fftw_real tmp50;
215 fftw_real tmp51;
216 ASSERT_ALIGNED_DOUBLE();
217 tmp33 = Y[-7 * iostride];
218 tmp34 = X[3 * iostride];
219 tmp35 = tmp33 + tmp34;
220 tmp74 = tmp33 - tmp34;
221 tmp50 = Y[-3 * iostride];
222 tmp51 = X[7 * iostride];
223 tmp52 = tmp50 - tmp51;
224 tmp75 = tmp50 + tmp51;
226 tmp36 = tmp32 + tmp35;
227 tmp57 = tmp32 - tmp35;
228 tmp53 = tmp49 + tmp52;
229 tmp61 = tmp52 - tmp49;
230 tmp73 = tmp71 - tmp72;
231 tmp83 = tmp71 + tmp72;
232 tmp76 = tmp74 - tmp75;
233 tmp84 = tmp74 + tmp75;
235 X[0] = tmp29 + tmp36;
236 Y[-7 * iostride] = tmp46 + tmp53;
238 fftw_real tmp38;
239 fftw_real tmp54;
240 fftw_real tmp37;
241 fftw_real tmp39;
242 ASSERT_ALIGNED_DOUBLE();
243 tmp38 = tmp29 - tmp36;
244 tmp54 = tmp46 - tmp53;
245 tmp37 = c_re(W[3]);
246 tmp39 = c_im(W[3]);
247 X[4 * iostride] = (tmp37 * tmp38) + (tmp39 * tmp54);
248 Y[-3 * iostride] = (tmp37 * tmp54) - (tmp39 * tmp38);
251 fftw_real tmp64;
252 fftw_real tmp66;
253 fftw_real tmp63;
254 fftw_real tmp65;
255 ASSERT_ALIGNED_DOUBLE();
256 tmp64 = tmp57 + tmp56;
257 tmp66 = tmp60 + tmp61;
258 tmp63 = c_re(W[1]);
259 tmp65 = c_im(W[1]);
260 Y[-5 * iostride] = (tmp63 * tmp64) - (tmp65 * tmp66);
261 X[2 * iostride] = (tmp65 * tmp64) + (tmp63 * tmp66);
264 fftw_real tmp58;
265 fftw_real tmp62;
266 fftw_real tmp55;
267 fftw_real tmp59;
268 ASSERT_ALIGNED_DOUBLE();
269 tmp58 = tmp56 - tmp57;
270 tmp62 = tmp60 - tmp61;
271 tmp55 = c_re(W[5]);
272 tmp59 = c_im(W[5]);
273 Y[-iostride] = (tmp55 * tmp58) - (tmp59 * tmp62);
274 X[6 * iostride] = (tmp59 * tmp58) + (tmp55 * tmp62);
277 fftw_real tmp94;
278 fftw_real tmp100;
279 fftw_real tmp98;
280 fftw_real tmp102;
281 fftw_real tmp93;
282 fftw_real tmp97;
283 ASSERT_ALIGNED_DOUBLE();
284 tmp93 = K707106781 * (tmp73 - tmp76);
285 tmp94 = tmp92 + tmp93;
286 tmp100 = tmp92 - tmp93;
287 tmp97 = K707106781 * (tmp83 + tmp84);
288 tmp98 = tmp96 - tmp97;
289 tmp102 = tmp96 + tmp97;
291 fftw_real tmp91;
292 fftw_real tmp95;
293 fftw_real tmp99;
294 fftw_real tmp101;
295 ASSERT_ALIGNED_DOUBLE();
296 tmp91 = c_re(W[2]);
297 tmp95 = c_im(W[2]);
298 Y[-4 * iostride] = (tmp91 * tmp94) - (tmp95 * tmp98);
299 X[3 * iostride] = (tmp95 * tmp94) + (tmp91 * tmp98);
300 tmp99 = c_re(W[6]);
301 tmp101 = c_im(W[6]);
302 Y[0] = (tmp99 * tmp100) - (tmp101 * tmp102);
303 X[7 * iostride] = (tmp101 * tmp100) + (tmp99 * tmp102);
307 fftw_real tmp78;
308 fftw_real tmp88;
309 fftw_real tmp86;
310 fftw_real tmp90;
311 fftw_real tmp77;
312 fftw_real tmp85;
313 ASSERT_ALIGNED_DOUBLE();
314 tmp77 = K707106781 * (tmp73 + tmp76);
315 tmp78 = tmp70 - tmp77;
316 tmp88 = tmp70 + tmp77;
317 tmp85 = K707106781 * (tmp83 - tmp84);
318 tmp86 = tmp82 - tmp85;
319 tmp90 = tmp82 + tmp85;
321 fftw_real tmp67;
322 fftw_real tmp79;
323 fftw_real tmp87;
324 fftw_real tmp89;
325 ASSERT_ALIGNED_DOUBLE();
326 tmp67 = c_re(W[4]);
327 tmp79 = c_im(W[4]);
328 X[5 * iostride] = (tmp67 * tmp78) + (tmp79 * tmp86);
329 Y[-2 * iostride] = (tmp67 * tmp86) - (tmp79 * tmp78);
330 tmp87 = c_re(W[0]);
331 tmp89 = c_im(W[0]);
332 X[iostride] = (tmp87 * tmp88) + (tmp89 * tmp90);
333 Y[-6 * iostride] = (tmp87 * tmp90) - (tmp89 * tmp88);
337 if (i == m) {
338 fftw_real tmp3;
339 fftw_real tmp7;
340 fftw_real tmp15;
341 fftw_real tmp20;
342 fftw_real tmp6;
343 fftw_real tmp12;
344 fftw_real tmp10;
345 fftw_real tmp21;
346 fftw_real tmp19;
347 fftw_real tmp22;
348 ASSERT_ALIGNED_DOUBLE();
350 fftw_real tmp1;
351 fftw_real tmp2;
352 fftw_real tmp13;
353 fftw_real tmp14;
354 ASSERT_ALIGNED_DOUBLE();
355 tmp1 = X[0];
356 tmp2 = X[3 * iostride];
357 tmp3 = tmp1 + tmp2;
358 tmp7 = tmp1 - tmp2;
359 tmp13 = Y[0];
360 tmp14 = Y[-3 * iostride];
361 tmp15 = tmp13 + tmp14;
362 tmp20 = tmp13 - tmp14;
365 fftw_real tmp4;
366 fftw_real tmp5;
367 fftw_real tmp8;
368 fftw_real tmp9;
369 ASSERT_ALIGNED_DOUBLE();
370 tmp4 = X[2 * iostride];
371 tmp5 = X[iostride];
372 tmp6 = tmp4 + tmp5;
373 tmp12 = tmp4 - tmp5;
374 tmp8 = Y[-2 * iostride];
375 tmp9 = Y[-iostride];
376 tmp10 = tmp8 + tmp9;
377 tmp21 = tmp8 - tmp9;
379 X[0] = K2_000000000 * (tmp3 + tmp6);
380 tmp19 = tmp3 - tmp6;
381 tmp22 = tmp20 - tmp21;
382 X[2 * iostride] = K1_414213562 * (tmp19 - tmp22);
383 X[6 * iostride] = -(K1_414213562 * (tmp19 + tmp22));
384 X[4 * iostride] = -(K2_000000000 * (tmp21 + tmp20));
386 fftw_real tmp11;
387 fftw_real tmp16;
388 fftw_real tmp17;
389 fftw_real tmp18;
390 ASSERT_ALIGNED_DOUBLE();
391 tmp11 = tmp7 - tmp10;
392 tmp16 = tmp12 + tmp15;
393 X[iostride] = (K1_847759065 * tmp11) - (K765366864 * tmp16);
394 X[5 * iostride] = -((K765366864 * tmp11) + (K1_847759065 * tmp16));
395 tmp17 = tmp7 + tmp10;
396 tmp18 = tmp15 - tmp12;
397 X[3 * iostride] = (K765366864 * tmp17) - (K1_847759065 * tmp18);
398 X[7 * iostride] = -((K1_847759065 * tmp17) + (K765366864 * tmp18));
403 static const int twiddle_order[] =
404 {1, 2, 3, 4, 5, 6, 7};
405 fftw_codelet_desc fftw_hc2hc_backward_8_desc =
407 "fftw_hc2hc_backward_8",
408 (void (*)()) fftw_hc2hc_backward_8,
410 FFTW_BACKWARD,
411 FFTW_HC2HC,
412 190,
414 twiddle_order,